http://www.diva-portal.org

This is the published version of a paper published in Frontiers in Psychology.

Citation for the original published paper (version of record):

Sioros, G., Miron, M., Davies, M., Gouyon, F., Madison, G. (2014) Syncopation creates the sensation of groove in synthesized music examples. Frontiers in Psychology, 5: 1036 http://dx.doi.org/10.3389/fpsyg.2014.01036

Access to the published version may require subscription.

N.B. When citing this work, cite the original published paper.

Copyright © 2014 Sioros, Miron, Davies, Gouyon and Madison. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Permanent link to this version: http://urn.kb.se/resolve?urn=urn:nbn:se:umu:diva-93059 ORIGINAL RESEARCH ARTICLE published: 16 September 2014 doi: 10.3389/fpsyg.2014.01036 Syncopation creates the sensation of groove in synthesized music examples

George Sioros 1, Marius Miron 2, Matthew Davies 2, Fabien Gouyon 2 and Guy Madison 3*

1 Department of Informatics Engineering, Faculdade de Engenharia, Universidade do Porto, Porto, Portugal 2 Institute for Systems and Computer Engineering - Technology and Science (INESC-TEC), Porto, Portugal 3 Department of Psychology, Umeå University, Umeå, Sweden

Edited by: In order to better understand the musical properties which elicit an increased sensation Sofia Dahl, Aalborg University of wanting to move when listening to music—groove—we investigate the effect of Copenhagen, Denmark adding syncopation to simple piano melodies, under the hypothesis that syncopation is Reviewed by: correlated to groove. Across two experiments we examine listeners’ experience of groove Maria Witek, University of Aarhus, Denmark to synthesized musical stimuli covering a range of syncopation levels and densities of Alan M. Wing, University of musical events, according to formal rules implemented by a computer algorithm that shifts Birmingham, UK musical events from strong to weak metrical positions. Results indicate that moderate *Correspondence: levels of syncopation lead to significantly higher groove ratings than melodies without any Guy Madison, Department of syncopation or with maximum possible syncopation. A comparison between the various Psychology, Umeå University, Mediagränd 14 A-123, Umeå, transformations and the way they were rated shows that there is no simple relation SE-901 87, Sweden between syncopation magnitude and groove. e-mail: [email protected] Keywords: groove, syncopation, movement, rhythm, listening experiment, music

INTRODUCTION increase groove. The latter was confirmed by a systematic analysis Certain types of music induce the desire in humans to tap their conducted by Davies et al. (2013) on simple rhythms. feet, move their body, and dance. This feeling is commonly known Madison and Sioros (2014) examined how musicians create as “groove” and has a strong affective component, as well as groove when they are confined to a monophonic melody and a strong correlation with music appreciation (Madison, 2006). an isochronous beat. Professional musicians were asked to per- According to Janata et al. (2012), we find pleasurable the music form six simple and six more complex melodies in two different that makes us want to dance and such music elicits spontaneous expressive ways: maximizing and minimizing groove. Their per- rhythmic movements. Furthermore, Madison et al. (2011) show formances were then rated by listeners to examine to what extent that the propensity to move differs substantially from one piece of the manipulations of the musicians created the intended effect. music to another, but can be linked to certain rhythmical proper- The study revealed that musicians tended to shift the onset and ties of the music signal such as beat salience and the density of offset positions of the notes in the simple melodies to faster met- events. rical levels in order to increase groove in their performances, and There is a strong agreement in the previous studies toward that this increased the groove to the same level as the complex the definition of groove. Madison (2006) defined groove as the melodies. By virtue of shifting note onsets and offsets to faster “sensation of wanting to move some part of the body in rela- metrical levels, the musicians created a greater number of off- tion to some aspect of the sound pattern.” Similarly, Janata et al. beat events, thus, increasing the syncopation. When they tried to (2012) conducted a survey involving a wide variety of music minimize groove they were unable to decrease it below the dead- related terms, to find that “groove is that aspect of the music pan level for the simple performances, but the dead-pan level that induces a pleasant sense of wanting to move along with the of groove in the complex melodies could be reduced to that of music.” Furthermore, there are strong clues that link certain prop- the simple melodies. This seemed to be achieved by two differ- erties of the music with the induced groove. Janata et al. (2012) ent means. One was to simplify the rhythmic structure toward conducted an extensive study on sensorimotor coupling which higher metrical levels, i.e., from 16th and 8th-notes to 8th, 4th, provides a broad analysis under the different facets of groove. The and half-notes, and the other was to break the rhythm such that computational part of the analysis suggests that music genre and the sense of a regular beat disappeared. Both of those devices faster tempi have a strong effect on the desire to move along to result in decreased syncopation but each in a different way. The the music. In comparison, Madison et al. (2011) claim that tempo first simply eliminates the syncopating notes by placing them on alone cannot explain groove. Consequently, Madison et al. (2011) the beat. In the second, syncopation disappears together with the found that low level descriptors, such as beat salience, event den- metrical feel. sity and fast metrical levels as physical attributes of the sound, While the results from Madison and Sioros (2014) appear to highly correlate with groove. In contrast and unlike what musico- indicate that the main physical correlate of groove was synco- logical literature points toward (Keil, 1995), microtiming does not pation, the design of the experiments was not fully controlled,

www.frontiersin.org September 2014 | Volume 5 | Article 1036 | 1 Sioros et al. Syncopation creates groove

since the musicians could, and did, change several aspects of their changing metrical positions only, without altering other expres- performances simultaneously. It is therefore difficult to determine sive or structural characteristics. The algorithm is an adaptation if the changes in some performance parameters were redundant, of the general syncopation transformations developed by Sioros independent, or interdependent, and also the unique effect of et al. (2013). While in the recent Witek study (2014)syncopa- each one of them. Therefore, there is a need to experimentally tion was measured in pre-existing drum loops, in our study we test the hypothesis that syncopation per se induces the sensation can generate and vary it in piano melodies using an automatic of groove independently of other expressive or structural features algorithm over which we have complete control. of the performances. We designed two listener experiments using simple piano Syncopation as a cognitive mechanism has been described as a melodies that were transformed in an automatic and controlled form of violation of metrical expectations (Huron, 2006, p. 297), way in order to generate syncopation without affecting other that is, temporal expectations that are established in the listener’s rhythmic or structural characteristics. In the first experiment we mind as a function of the temporal regularities in the music sig- explore the more general hypothesis that increasing the synco- nal. Several operational definitions of syncopation are based on pation to a moderate amount and introducing faster metrical the more general expectation principle, and provide algorithms levels results in an increased groove sensation. Then, in the sec- for measuring the amount of syncopation present in a rhyth- ond experiment, we aim to gain more insight on how the strength mic pattern (Longuet-Higgins and Lee, 1984; Gómez et al., 2005; of each individual syncopation felt in a melody influences groove Huron and Ommen, 2006; Sioros and Guedes, 2011). A common as well as distinguishing between different ways of introducing characteristic among many syncopation measures is the use of a faster metrical levels. Parallel to the syncopation transformations, metrical template that describes the strong and weak beats found density transformations that increase the number of events in in the musical meter. Normally, a note articulated in a weak met- the melodies are used as controls in order to verify the effect of rical position is followed by a note in the following strong position syncopation. resulting in a weak-strong rhythmic figure. Huron (2006, p. 295) Basedonrecentresultsonthistopic(Madison and Sioros, suggests that when the expected note on the strong position does 2014; Witek et al., 2014), our predictions for the outcome of not occur, the tension increases and the feeling of syncopation the listening experiments are: (1) the deadpan versions of the is evoked. The syncopation is then attributed to the “improper melodies that contain no syncopation would be rated lowest in binding” of a musical event found on a weak metrical position groove, (2) introducing a small number of strongly syncopating to an event in the following strong position, which is delayed or notes would be rated highest in groove, (3) an increased num- not present at all (see also London, 2012, p. 107). Different kinds ber of syncopated notes would not have a proportional increase of weightings related to the strength of the metrical positions have in groove ratings or would be rated lower. been used in order to quantify syncopation (see Gómez et al., 2007 for various syncopation metrics). EXPERIMENT 1 In the context of syncopation and its relation to groove, a METHODS very recently published study (Witek et al., 2014) demonstrated Participants a listener preference for moderate syncopation in drum pat- Twenty-eight participants (14 female, 14 male, mean = 30.0 terns. The study was made using funk drum-breaks with varying years, SD = 2.9 years) of different nationalities were recruited degrees of syncopation that were taken in their majority from real through advertisement on Umeå university campus in Sweden music tracks (36 out of 50). The degree of syncopation was then and INESC-TEC in Portugal. They received a remuneration for measured using a variant of the algorithm proposed by Longuet- their participation of 75 Kronor at Umeå university or 10Cat Higgins and Lee (1984). As syncopation has been often associated INESC-TEC. with rhythmic tension and rhythm complexity (Huron, 2006, p. 295), the study concluded that gradually increasing the synco- Stimuli pation also increases the desire to move, but only up to a certain Our aim was to measure and understand the effect of syncopation point beyond which desire decreases. on the sensation of groove, independently of other structural or Additionally, the pleasure experienced when listening to music expressive factors such as the meter or timing deviations from the is often correlated with the violation of expectations (Huron, metronomical positions. At the same time we wanted to eliminate 2006, p. 297). As a component of musical rhythm, syncopation any expressive features of human performances, such as timing is a way to create metrical tension (Temperley, 1999), owing to deviations from the metrical grid or dynamic accents. To this end, the fact that it occurs only in contradiction with an established we used simple piano melodies with a clear metrical structure. All metrical pattern (Longuet-Higgins and Lee, 1984), thus breaking melodies were in 4/4 meter and consisted of short melodic phrases an existing expectation. Correspondingly, moving from very sim- of2–4barsasintheexampleofFigure 1. The melodies contain ple to more intricate rhythm patterns can elicit different kind of simple rhythmic figures such as the ones found in melodies for responses from people in terms of enjoyment and desire to move. children and were of moderate tempo (120 BPM). In this paper, we focus on syncopation as the physical prop- Seven melodies were traditional songs for children from dif- erty of the music signal that can drive the perception of groove. ferent cultures. Another five were composed for the purposes of As we want to eliminate all other expressive factors of a musical the experiment. The traditional melodies had a more complex performance, we developed a computer algorithm that gener- structure compared to the composed ones, with higher variety ates syncopation in monophonic sound sequences relying on of phrase lengths, phrase boundaries positions and harmonic and

Frontiers in Psychology | Auditory Cognitive Neuroscience September 2014 | Volume 5 | Article 1036 | 2 Sioros et al. Syncopation creates groove

FIGURE 1 | Example of a piano melody and all the applied anticipated by a 16 note. The “weak” quarter notes are shifted to the 8 note transformations of experiment 1. The beats are annotated above the staffs. position preceding the following strong beat. (D) 16D8A. The last “strong” (A) The original quantized melody. An arch marks the duration of each phrase note of each phrase is anticipated by an 8 note. The “weak” quarter notes of the melody (in this example: the first two bars and the last two bars). (B) are shifted to the 16 note position preceding the following strong beat. (E) 8D8A. The last “strong” note of each phrase is anticipated by an 8 note. The 16D16A. The last “strong” note of each phrase is anticipated by a 16 note. “weak” quarter notes are shifted to the 8 note position preceding the The “weak” quarter notes are shifted to the 16 note position preceding the following strong beat. (C) 8D16A. The last “strong” note of each phrase is following strong beat. melodic content. A set of simple transformations was then applied is a half or whole note articulated on a strong beat, as a direct to them resulting in 5 total versions for each. consequence of the way the melodic phrases were segmented. The The original melodies included only metrical levels equal to or anticipated offbeat notes are tied to the following silent strong slower than the quarter note, i.e., quarter notes (q.n. = 500 ms), beats where they originally belonged giving rise to the feeling of half notes (1000 ms) and whole notes (2000 ms). They contained syncopation. Therefore, the expectation of a note onset occurring no syncopation (Figure 1, staff A) and were all quantized so that at the strong beat is violated and the established meter is momen- their timing did not deviate from the metronomic positions. To tarily contradicted (Temperley, 1999; Huron, 2006, p. 295; Huron each melody, we applied a set of transformations in order to and Ommen, 2006). introduce syncopation (Figure 1, staffs B–E). While the original The syncopation transformations aim to introduce syncopa- version contained only slower metrical levels, applying any of the tion in a way that the perceived meter and other rhythmical and transformations introduced the 8 note metrical level (250 ms) structural characteristics of the melodies are preserved. In order and/or the 16 note metrical level (125 ms). not to affect the perceived meter, the remainder of the notes on Before applying any transformation, the melodies were man- strong beats, i.e., all notes on the first and third quarter note of ually segmented into short phrases of 2–4 bars. To this end, each bar besides the last one of each phrase, were kept in their the following procedures were applied. The boundaries of the original positions. However, the notes found in the weak posi- melodic phrases were chosen according to note durations. When, tions, i.e., on the second and fourth quarter note, were delayed so after a long note, i.e. a half or whole note, a series of quarter notes that the previous note, articulated on a strong beat, is now length- begins, the first quarter note marks the beginning of a phrase. ened and therefore felt as accented (Povel and Okkerman, 1981). While this rule would be insufficient to define phrase bound- In this way, the meter is first established by the notes on the strong aries in more complex melodies, it is valid and efficient for the beats at the beginning of each phrase. Then, the anticipated note simply structured phrases of the music examples (MEs) used in is placed at an offbeat weak position and is tied to the follow- this experiment. The transformations were then applied on each ing strong beat where it originally belonged (Figure 1, staffs B–E) phrase separately. generating syncopation. The note articulated in the weak posi- The MIDI files of all melodies and their respective variations tion before the last note of each phrase is not delayed in order to are available as Supplementary material. leave space for the syncopating note to be anticipated. The delay of the weakly articulated notes introduces faster metrical levels in Syncopation transformations. The transformations employ the addition to the metrical levels introduced by the syncopation. principles of the “improper binding” of notes articulated in weak Four different syncopation transformations were applied on metrical positions to missing notes in the following strong posi- the original melodies that differ on the metrical subdivisions that tions in order to generate syncopation (London, 2012, p. 107). the displaced notes were found after the transformation. They are First, the meter is divided in strong and weak beats. Then, the coded by two numbers, x and y,inthefollowingformxDyA, last note event of each phrase is syncopated by anticipating it to where x denotes the metrical level that the weak quarter notes an earlier weaker metrical position. The last note of each phrase were shifted to and y is the metrical level where the last note of

www.frontiersin.org September 2014 | Volume 5 | Article 1036 | 3 Sioros et al. Syncopation creates groove

each phrase is shifted to. For example, in the 16D8A transforma- rated two music examples which allowed them to set the volume tion the notes originally articulated on weak quarter notes were to a comfortable level and to familiarize themselves with the pro- shifted to the 16 note position directly preceding the next strong gram and the procedure. The sliders were locked while the music beat, while the last note of each phrase in the melodies was shifted was playing to force the participants to listen to the entire exam- to the preceding 8 note metrical position (Figure 1, staff D). ple before rating. After rating each example the participant could continue to the next by clinking on a button. The entire session Implementation of musical examples lasted between 30 and 40 min. To generate all the versions for each melody, the original melodies were imported as MIDI files in Ableton Live where they were RESULTS quantized and edited in order not to contain durations shorter Since there were a different number of melodies in each group, than quarter notes. Simple Max for Live devices were developed in this precluded a full-factorial model. A within-participants 5 order to automatically apply the corresponding transformations. transformations × 5 melody 2-Way ANOVA demonstrated main MIDI files were generated for all 6 versions of each melody and effects for the simple melodies of transform [F(4, 84) = 6.678, were then rendered into 16 bit wave files using high quality piano p < 0.0005] and melody [F(4, 84) = 11.56, p < 0.000001], but 1 samples . not of their interaction [F(16, 336) = 1.604, p = 0.066]. The com- plex melodies exhibited a similar pattern of effects, according Rating scales to a 5 transformations × 7 melody 2-Way ANOVA. There was The experiment was intended to explore the “movement induc- amaineffectoftransform[F(4, 84) = 5.426, p < 0.001] and ing” effect of syncopation. To this end, participants were asked to melody [F(6, 128) = 12.55, p < 0.000001], but not of their inter- rate groove while listening to musical stimuli. Groove was defined action [F(24, 504) = 1.060, p = 0.39]. Figure 2 gives an overall as “the sensation of wanting to move some part of your body in impression about the range of groove ratings across melodies and relation to some aspect of the music.” However, the participants the size of the trends for the transforms, where melodies 1–5 are were told that the purpose of the experiment was the general study the simple ones and 6–12 the complex traditional songs. It can be of rhythm perception, so that the response bias was minimized. seen that the original quantized versions always have the lowest For this reason two additional rating scales were included in the ratings, and that 16D8A and 16D16A transformations are most experiment, namely familiarity and preference. often at the top. The post-hoc p-values in the above analysis were All three scales were presented to participants under the global p < 0.05. question “How well do the following words describe your experi- There are some differences in magnitude between the trans- ence of the music?” The ratings ranged from 0 for “not at all” to 10 forms for the two types of melodies, as indicated by the confi- for “entirely” in unit increments. A separate horizontal slider ini- dence intervals in Figure 3. Significant differences in the groove tially positioned at 5 was available for entering the ratings for each ratings can be seen between the transformations where the mean scale. The scales were presented in the same order throughout the valueofgrooveofonetransformationdoesnotoverlapwith experiment. the confidence interval of another transformation. The simple melodies generally have lower ratings, but they are increased more Design by the transformations than the complex melodies. For the trans- The dependent variable was the rating of groove, and the inde- formed versions, there are only small differences between the pendent variables were (1) the type of transformation of the melodies (5 levels including 1 quantized and 4 different types of syncopation), (2) melody type (simple or complex), and (3) the melodies themselves. A within-participants design was employed in which each participant rated all 60 conditions (5 transforms × 12 melodies), which were presented in a different random order to each participant. Randomization and other functionality was built into a custom-made presentation and response recording software.

Procedure The participants listened to the music examples alone through headphones in a quiet room and rated them on a computer. First, a paper with instructions that included the definitions of the rating scales was given to them. Prior to the beginning of the listening test, they were asked if the task and terminology was clear. After signing the informed consent form, a training ses- sion preceded the experiment proper, in which the participants FIGURE 2 | Mean groove ratings for transform and melody, across participants. Melodies 1–5 are the simple ones and melodies 6–12 are the 1 . . http://www native-instruments com/en/products/komplete/samplers/ complex traditional songs. kontakt-5/overview/library/

Frontiers in Psychology | Auditory Cognitive Neuroscience September 2014 | Volume 5 | Article 1036 | 4 Sioros et al. Syncopation creates groove

detail how syncopation affects the sensation of groove. In order to be able to achieve a higher degree of syncopation without alter- ing the perceived meter, a metronome was employed to provide a strong metrical reference.

Participants Twenty-two participants (6 female, 16 male) took part in the experiment (mean = 30.7 years, SD = 6.3 years). Of the 22 par- ticipants, six of them had no music training and nine of them were considered professional musicians with more than 8 years of music training. The participants were recruited via email, did not participate in experiment 1 and did not receive any remuneration for their participation

Stimuli Five of the traditional complex songs from Experiment 1 and all five simple melodies that were composed for the purposes FIGURE 3 | Mean groove ratings for transform and melody type across of Experiment 1 were used. A set of transformations was then participants. applied resulting in 6 versions of each. The meter was emphasized by a hi-hat drum sound on every quarter note with the first quar- Table 1 | Effects sizes for each transform relative to Quantized, ter note of each bar being dynamically stressed (twice the MIDI separate for simple and complex melodies. velocity value). An introductory bar with only the metronome sound preceded the playback of each melody. Transform Melody type We wanted to compare the effect of the syncopation on the Simple N = 110 Complex N = 154 sensation of groove against the effect of transformations that did not introduce syncopation and preserved the main rhythmical 8D8A 0.295 0.221 and structural characteristics of the original melodies like the 8D16A 0.228 0.204 meter and phrase boundaries. To this end, we applied two kinds 16D8A 0.306 0.283 of transformations: (1) shifting note onset positions to introduce 16D8A 0.347 0.337 syncopation (Figure 4, staffs B,C,D) and (2) density transforma- tionsthatdoubledthenumberofnotesperbarinthemelodies All differences are significant (t = 2.71–4.19, df = 218/301, p < 0.01). without introducing syncopation (Figure 4, staffs E,F). In total, 6 different versions of each melody were generated: the origi- simple and complex melodies. The only difference that seems nal, three syncopated versions and two with increased density. to be significant is that for the complex melodies the 16DxA Before applying any syncopation transformations, the melodies note transformations are somewhat more effective than the 8DxA were manually segmented into short phrases of 2–4 bars as in transformations. Experiment 1. The transformations were then applied on each Table 1 lists the effect sizes (Cohen, 1988) for each transforma- melodic phrase separately. tion and melody type, which shows that Cohen’s d varies between While the original version of each melody contained only 0.2 and 0.35. slower metrical levels, i.e., only quarter notes (q.n. = 500 ms), half notes (1000 ms) and whole notes (2000 ms), applying any EXPERIMENT 2 of the transformations introduced the faster 8th note metrical METHODS level (250 ms) but no 16th notes. A detailed description of the Experiment 1 showed that the syncopation transformations transformations follows. increased the groove ratings of the simple melodies, but also that The MIDI files of all melodies and their respective variations all transformations had a similar effect. This raises the question are available as Supplementary material. whether any kind of transformation that introduces faster met- rical levels but preserves the structure of the melodies would in Syncopation transformation 8A. The 8A transformation intro- fact result in higher ratings from the metronomic deadpan origi- duced a syncopating event at the end of each phrase (Figure 4B) nal version, or if syncopation is indeed required. Such an example by anticipating the last note of each phrase by an 8 note. Similar would be a transformation of the original quantized melody that to the transformations of Experiment 1 the perceived meter was preserves the order and pitch of the note events as well as the not affected since the remainder of the notes on strong beats, i.e., original meter and phrase boundaries (see Figure 4, staffs E,F). all notes on the first and third quarter note of each bar besides the In Experiment 2 we aim to address this main question by last one of each phrase, were kept in their original positions. In including other non-syncopation transformations. addition to this end phrase anticipation, we generated additional Additionally, we employ a set of three syncopation transfor- syncopation by anticipating the notes found on the weak beats in mations that vary in the degree and strength of the syncopation each phrase, i.e., on the second and fourth quarter note of each that they generate. We wanted, in that way, to examine in more bar, thus not affecting the perceived meter. Those syncopations

www.frontiersin.org September 2014 | Volume 5 | Article 1036 | 5 Sioros et al. Syncopation creates groove

FIGURE 4 | Example of a piano melody and all the applied (C) Transformation 8D. The last “strong” note of each phrase is anticipated transformations of experiment 2. The beats are annotated above the staffs. by an 8 note. The “weak” quarter notes are delayed by an 8 note. (D) (A) The original quantized melody. An arch marks the duration of each phrase Maximum syncopation. All notes have been anticipated by an 8 note. (E) of the melody (in this example: the first two bars and the last two bars). (B) Density1. All note durations have been divided in two thus duplicating each Transformation 8A. The last “strong” note of each phrase is anticipated by an note. (F) Density2. Before each note of the original melody, a new note of the 8 note. The “weak” quarter notes are anticipated by an 8 note. same pitch is introduced. are felt less strongly since they involve faster and weaker metri- Density 2. The density 2 transformation doubled the number of cal levels (Longuet-Higgins and Lee, 1984), i.e., the quarter note note events by introducing an 8 note before each note in the orig- metrical subdivision instead of the half bar or the bar. inal melody (Figure 1, staff F).Theintroducednotehadthesame pitch as the original note following it. One can think of the trans- Syncopation transformation 8D. The 8D transformation is iden- formation as the superposition of the original melody and the tical to the 8D8A transformation of Experiment 1. maximum syncopation transformation. Just as with the density 1 transformation, no syncopation was introduced. Maximum syncopation. The aim of the third syncopation trans- Rating scales formation was to create a greater amount of syncopation. To this The rating scales were identical to those used in Experiment 1. end, all notes in each original melody were anticipated by an 8 note. Although such a transformation shifts the entire metrical Design perception, syncopation was produced through the phase offset The dependent variable was the rating of groove and the between the melody and the metronome. independent variables were (1) the type of transforms of the melodies (6 levels including 1 quantized, 2 density, and 3 syn- Density transformations. Two density transformations were copation), (2) melody type (simple or complex), and (3) the applied. In density 1, new notes were introduced after the orig- melodies themselves. A mixed design was employed such that inal ones (Figure 1, staff E), and in density 2,newnoteswere two groups of participants rated half the conditions each, in introduced before the original ones (Figure 1, staff F). order to make the task less taxing. Group 1 were given simple melodies #1 and #2 and the complex melodies #6, #7, and #8, Density 1. The density transformations doubled the density of and group 2 were given simple melodies #3, #4, and #5 and note events per bar by introducing new notes of the same pitch the complex melodies #9 and #10. The 30 within-participants as the original notes. The density 1 transformation introduced conditions (5 melodies × 6 transforms) were presented in a after each note event in the original melody a new one of the different random order to each participant. Participants were same pitch, effectively halving each duration (Figure 1, staff E). required to listen to the whole recording before entering their For example, each quarter note became two 8 notes, each half ratings. note became two quarter notes etc. Since the original melodies contained no syncopation, the density transformation did not Implementation of the music examples introduce syncopation, because all newly introduced notes on The implementation of the music examples was identical to those weak metrical positions were followed by the original notes of the in Experiment 1, except for the addition of the metronome melodies articulated in the strong metrical positions. backing.

Frontiers in Psychology | Auditory Cognitive Neuroscience September 2014 | Volume 5 | Article 1036 | 6 Sioros et al. Syncopation creates groove

Procedure Melody interaction [F(20, 380) = 1.964, p < 0.01]. In addition, The experiment was conducted online using a web-based inter- the Transformations x Melody interaction was also significant face, first developed for use in Davies et al. (2013).Theuser [F(20, 380) = 2.418, p < 0.001]. Melody did not relate to any lin- interface provided instructions for the participants to take the ear dimension, but was included to pick up whatever systematic experiment in one session in a quiet listening environment, using variance that happened to concur with the numbering of the either high quality headphones or loudspeakers. After reading the melodies. Figure 5 shows that the two sparse syncopation trans- instructions and in order to proceed to the experiment, the par- formations were the most effective. Significant differences in ticipants consented to take part in the experiment. In addition, the groove ratings are indicated by the confidence intervals in they entered some personal information including their sex, age Figure 5. For example, the max syncopation transformation has and the number of years of musical training. a significant difference in the groove ratings from all other trans- To allow participants to set the playback volume to a com- formations,whilethemeanratingofthedensity 1 transformation fortable level, become familiar with the type of stimuli and to is not significantly different from that of density 2.Thepost-hoc acquaint themselves with the functionality of the interface, a short p-values in the above analysis were p < 0.05 training phase was conducted prior to the start of the main exper- Effect sizes for the differences between quantized and the iment. After the training phase, participants proceeded to the transformations were quite respectable, and amounted to 0.208 main experiment, during which their ratings were recorded. To for Density 1, 0.233 for Density 2, 0.439 for MaxSync, 0.601 for reduce the time needed for participants to complete the exper- 8D, and 0.597 for 8A. iment, the stimuli were split into two groups of equal size as described in Section Design. Upon completion of the main exper- DISCUSSION iment, participants were able to offer feedback and comments The purpose of the current study was to examine the effect which were recorded along with their ratings. of syncopation on the perception of groove independently of other expressive features of music performances. To this end, RESULTS pianomelodieswereusedthatweretransformedbyanalgo- As described before, the melodies were distributed among two rithm designed to introduce syncopation in a systematic and groups of participants so as to reduce the workload of each indi- automatic way. We hypothesized that the syncopation transfor- vidual. This division was made to diffuse possible group effects, mations would result in higher groove ratings than the deadpan rather than to examine group x melody interactions. Since there versions of the melodies with no syncopation. Additionally, we were 5 simple and 5 complex melodies, each of these sets was predicted that a higher number of introduced instances of syn- therefore broken up in 2 and 3 melodies that were assigned to each copation would not have a proportional effect in the perception group as described in Section Design. A mixed model 5 melody of groove and that the relation between syncopation and groove × 6 transformations within × 2 listener group between 3-Way was complex—in the sense that it is not simply the amount of ANOVA was initially conducted to examine possible effects of syncopation, but how it is expressed that causes the increase in group. Melody type (simple vs. complex) was excluded from the the sensation of groove. In the simplest form this could follow ANOVA model because it rendered incomplete cells and unequal an inverted U shape although an understanding of the relation cell sizes. There were main effects of group [F(1, 38) = 4.20, p = 0.047] and transform [F(5, 190) = 22.74, p < 0.000001], but no interaction effects, notably not for any one involving group, such as group × melody [F(3, 152) = 1.25, p = 0.293], group × transform [F(5, 190) = 1.37, p = 0.237], or group × melody × transform [F(20, 160) = 1.03, p = 0.422]. Thus, the only signif- icant effect of group was that ratings were generally higher in group 1 (2.505 vs. 1.743, N = 1200, pooled SD = 2.186). Since this effect cannot alter the effect of the transform in any appre- ciable way, the data from both groups were aggregated. This means that we cannot determine to what extent the group dif- ference emanated from the participants or the melodies, but this is of no consequence as melodies were not part of any hypoth- esis. A within-participants 2 Melody type × 5Melody× 6 Transformations 3-Way ANOVA was conducted with groove as the dependent variable. The unit of analysis were pairs of par- ticipants, since melodies were originally distributed across two groups, and were entered as a random-effects variable. Main effects were found for Transformations [F(5, 95) = 5.459, p < FIGURE 5 | Mean groove ratings for each transform, across 0.0005] and Melody [F(4, 76) = 97.37, p < 0.000001], but not for participants and melodies (including melody type). Effect sizes for the differences between quantized and the syncopated transforms were quite Melody type [F(1, 19) = 0.686, p = 0.42]. As expected from the respectable, and amounted to 0.208 for Density 1, 0.233 for Density 2, small Melody type effect, no effects involving this variable were 0.439 for MaxSync, 0.601 for 8D, and 0.597 for 8A. significant except the 3-Way Transformations × Melody type ×

www.frontiersin.org September 2014 | Volume 5 | Article 1036 | 7 Sioros et al. Syncopation creates groove

between syncopation and groove requires a closer examination of 2005)aswellaswiththerecentstudyofSong et al. (2013) that the characteristics of individual instances of syncopation. All our points out the importance for the felt syncopation of the loca- hypotheses were verified in the experiments. tion of the individual instances of syncopation in relation to the In Experiment 1, all transformations generated syncopation meter. At the same time their position in each phrase results in by anticipating the last note of each phrase in the melodies at grouping them together with the previous strong beat creating a the same time that they introduced faster metrical levels. The Short–Long rhythmic figure with the short duration on the beat. transformations displaced notes from the slower metrical levels, The same kind of rhythmic figure is found in the 8D variation quarter note or half bar levels, to the faster 8th or 16th note levels but with the long note on the beat, resulting in no syncopation. generating a variety of combinations of note durations. However, This rhythmic figure repeats at least three times in most phrases this variety was not reflected in the groove ratings which were before the final strong syncopation arrives. Together with the increased at similar levels for all the transformations. metronome and the harmonic structure of the melodies, these In Experiment 2, we verified that the increase in the groove repetitions may enforce the meter, rather than contradict it, in ratings was in fact particular to the syncopation, and that other bothversions.Groovemaythereforebeaffectedonlybythesyn- types of transformations tended to have weaker effects on groove. copating notes that clearly contradict our metrical expectation at In addition, Experiment 2 examined whether the strength of metrically strong positions at the beat level and not by the weaker the syncopation had an effect on groove. Our hypothesis was that syncopation at the subdivision of the beat. a moderate number of syncopating notes that contradicted the On the other hand, in the maximum syncopation,syncopation meter only momentarily would increase the felt groove, while, comes about through the interaction between the metronome in contrast, a high number of syncopating notes that constantly and the melody. The melody, if it was to be heard alone, contains challenged the meter would not contribute to the groove percep- no syncopation. In this case, the syncopation cannot be attributed tion. The hypothesis was tested by using two different strategies to individual syncopating events but to the interaction of rhyth- in generating syncopation. The first syncopation transformations, mic layers during the entire duration of the music, resulting in an 8A and 8D transformations, syncopated certain notes while keep- overall feeling of syncopation with no distinct moments. ing most of the notes in place on the strong beats. In contrast, However, if the listeners ignored the metronome sound they the maximum syncopation transformation displaced the entire could have perceived the melodies as non-syncopating and identi- melody and created a constant phase offset between the melody cal to the original versions. Although this would explain the lower and the metronome sound. Both strategies resulted in higher ratings compared to the other syncopating versions, it would raise groove ratings compared to the original versions, but less so for new questions about the higher ratings compared to the original maximum syncopation. versions or to those of the two density transformations. At the Conversely, the 8A and 8D transformations did not differ in same time, the metronome sound with an emphasized downbeat their respective groove ratings although they differ in the num- was heard for one bar alone before each melody. On this basis, it ber of syncopating notes they generated. While 8D generated seems unlikely that the listeners would ignore it the moment the only a single syncopating note at the strong beat at the end of syncopating melodies began. Then, even if, in the course of hear- each phrase, 8A syncopated a substantial number of additional ing the melody, they shifted their perception to align the meter notes found on weak beats and that were simply delayed in 8D. to it, the contradiction with the stable metronome would per- One possible explanation for this result is that the effect is due sist, thus evoking a constant feeling of syncopation together with to the introduction of faster metrical levels and not specific to a strong sense of a stable meter and beat. the syncopation. However, the two density transformations— Hence, the ratings show that a moderate amount of syncopa- which introduce the same faster metrical levels as the syncopation tion arising from notable salient instances of syncopation that transformations without generating any syncopation—were rated underline the melodic boundaries contribute more to the per- significantly lower than the syncopation transformations. ception of groove than a higher degree of syncopation that is Another explanation of the above results considers the nature uniformly distributed along the entire duration and does not of the syncopation in each transformation. In 8A and 8D,the relate to the melodic structure. Most noticeably, the weakening of meter is first clearly established and a strong beat sensation is cre- the groove sensation when the amount of syncopation is increased ated at the half bar metrical level. Then, at the end of each phrase, isnotduetoabreakdownoftheperceivedmeter. a single syncopating event violates our strong expectation that The results are compatible with the recent study by Witek et al. the last note should occur on the pre-established beat. The syn- (2014) in which a moderate degree of syncopation was preferred copation is felt very strongly because of the well-established beat toahighdegreeofsyncopation.However,ourcurrentstudydif- and the position of syncopation in relation to the structure of the fers significantly from Witek et al.’s. The synthesized stimuli were melody. quite different in the two studies. Witek et al. used a number The “extra” syncopation in the 8A version that results from of real drum-breaks containing several timbres and each differ- the anticipation of the notes articulated on the weak quarter note ent example had a different degree of syncopation. In the present level, i.e., on the second and fourth quarter note positions, is felt study, monophonic piano melodies that originally contained no less strongly. The weak quarter note level comprises a subdivision syncopation were used, and syncopation was systematically gener- of the more salient half bar pulse, which means that our expecta- ated by a computer algorithm resulting into different syncopated tion about events on these positions is accordingly lower. This is versions of the same melody. The simple rhythmic structure of the in accordance with several metrics of syncopation (Gómez et al., original melodies allowed for applying deterministic algorithms

Frontiers in Psychology | Auditory Cognitive Neuroscience September 2014 | Volume 5 | Article 1036 | 8 Sioros et al. Syncopation creates groove

to manipulate their rhythmic characteristics. Introducing synco- Our maximum syncopation condition, even though it increased pation in already complex music could have a strong effect on the the syncopation did not cause the breakdown of the perceived perceived meter, such as a shift of the downbeat. The automatic meter despite constantly challenging it. Furthermore, a richer transformations of the experiments affected the particular rhyth- rhythmic structure in note durations and metrical levels was mic properties they were designed to change exclusively, that is, considered a possible explanation for the relation between syn- syncopation and density. More importantly, this method allowed copation and groove as syncopation is inherently related to such us to draw specific conclusions about how syncopation can be structures. However, no definite conclusion was possible based on applied effectively in order to create groove. the results of the experiments. The syncopation transformations have the secondary effect of increasing the number of different note durations in the ACKNOWLEDGMENTS melodies. This is a direct consequence of their design. Excluding This research was supported by grant P2008:0887 from the the final notes in each melody, the original melodies contained Bank of Sweden Tercentenary Foundation awarded to Guy only two durations, which can be thought of as long (half notes) Madison and by the CASA project with reference PTDC/EIA- and short (quarter notes). The syncopation transformations in CCO/111050/2009 (Fundação para a Ciência e a Tecnologia). both experiments, with the exception of the maximum syncopa- tion transformation, enrich the rhythms by introducing a variety of shorter and in-between short and long durations. The density SUPPLEMENTARY MATERIAL and the maximum syncopation transformations do not have this The Supplementary Material for this article can be found . . . . effect. Although they all shorten the durations of the notes in online at: http://www frontiersin org/journal/10 3389/fpsyg . the original melodies, they produce simple short–long relations. 2014 01036/abstract The result of Experiment 2 that the maximum syncopation is rated lower than the other two syncopated versions suggests REFERENCES that a variety in note durations could have a contribution to Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences. Hillsdade, NJ: the sensation of groove. The transformations of experiment 1 Routledge. introduce the various note durations gradually with the 8D16A Davies, M., Madison, G., Silva, P., and Gouyon, F. (2013). The effect of microtim- and 16D8A resulting in higher variety than the 8D8A and ing deviations on the perception of groove in short rhythms. Music Percept. 30, 497–510. doi: 10.1525/mp.2013.30.5.497 16D16A, due to the fact that they mix the 16 and 8 note metrical Gómez, F., Melvin, A., Rappaport, D., and Toussaint, G. T. (2005). “Mathematical subdivisions. However, this is not reflected in the groove ratings measures of syncopation,” in Proceedings of the BRIDGES: Mathematical that would be expected to be higher for the 8D16A and 16D8A if Connections in Art, Music and Science (Banff, AB), 73–84. a direct relation between the variety in note durations and groove Gómez, F., Thul, E., and Toussaint, G. T. (2007). “An experimental comparison of perception existed. Furthermore, despite the density 1 trans- formal measures of rhythmic syncopation,” in Proceedings of the International Computer Music Conference (Copenhagen), 101–104. formation generating greater number of durations and faster Huron, D. (2006). Sweet Anticipation: Music and the Psychology of Expectation. metrical levels compared to maximal syncopation—although Cambridge, MA; London: The MIT Press. much less compared to the 8D or 8A—, maximal syncopation Huron, D., and Ommen, A. (2006). An empirical study of syncopation in was rated higher than density 1. American popular music, 1890-1939. Music Theory Spectr. 28, 211–231. doi: 10.1525/mts.2006.28.2.211 The relation between the diversity in note durations or the Janata, P., Tomic, S. T., and Haberman, J. M. (2012). Sensorimotor coupling in articulation of faster metrical levels in rhythms and groove per- music and the psychology of the groove. J. Exp. Psychol. Gen. 141, 54–75. doi: ception therefore needs to be further explored as the results of 10.1037/a0024208 the current study are inconclusive on this question. Nevertheless, Keil, C. (1995). The theory of participatory discrepancies: a progress report. the present study provides a systematic method for designing Ethnomusicology 39, 1–19. doi: 10.2307/852198 London, J. (2012). Hearing in Time. 2nd Edn. New York, NY: Oxford University algorithms that can produce versions of melodies that contain Press. doi: 10.1093/acprof:oso/9780199744374.001.0001 faster and slower metrical levels as well as various degrees and Longuet-Higgins, H. C., and Lee, C. S. (1984). The rhythmic interpretation of diversity of note durations and different distributions of notes monophonic music. Music Percept. 1, 424–441. doi: 10.2307/40285271 among those durations. It provides a method for directly com- Madison, G. (2006). Experiencing groove induced by music: consistency and paring the different versions and examining them against listener phenomenology. Music Percept. 24, 201–208. doi: 10.1525/mp.2006.24.2.201 Madison, G., Gouyon, F., Ullén, F., and Hörnström, K. (2011). Modeling the ratings. tendency for music to induce movement in humans: first correlations with In conclusion, syncopation alone was sufficient to increase low-level audio descriptors across music genres. J. Exp. Psychol. Hum. Percept. groove in simple melodies. We confirmed that a moderate Perform. 37, 1578–1594. doi: 10.1037/a0024323 amount of syncopation is more effective in creating the sensa- Madison, G., and Sioros, G. (2014). What musicians do to induce the sensation of tion of groove than a very high amount. However, the amount of groove in simple and complex melodies, and how listeners perceive it. Front. Psychol. 5:894. doi: 10.3389/fpsyg.2014.00894 syncopation, as a scalar quantity, that characterized each exam- Povel, D.-J., and Okkerman, H. (1981). Accents in equitone sequences. Percept. ple was not sufficient to explain how syncopation increased the Psychophys. 30, 565–572. doi: 10.3758/BF03202011 sensation of groove. To further explore this relation, an analysis Sioros, G., and Guedes, C. (2011). “Complexity driven recombination of MIDI of the individual instances of syncopation in the music examples Loops,” in Proceedings of the 12th International Society for Music Information Retrieval Conference (Miami, FL), 381–386. was needed. This analysis showed that individual instances of syn- Sioros, G., Miron, M., Cocharro, D., Guedes, C., and Gouyon, F. (2013). copation that momentarily created rhythmic tension were more “Syncopalooza: manipulating the syncopation in rhythmic perfor- important in groove than an overall higher degree of syncopation. mances,” in Proceedings of the 10th International Symposium on Computer

www.frontiersin.org September 2014 | Volume 5 | Article 1036 | 9 Sioros et al. Syncopation creates groove

Music Multidisciplinary Research (Marseille: Laboratoire de Mécanique et Received: 01 June 2014; accepted: 29 August 2014; published online: 16 September d’Acoustique), 454–469. 2014. Song,C.,Simpson,A.J.R.,Harte,C.A.,Pearce,M.T.,andSandler,M.B. Citation: Sioros G, Miron M, Davies M, Gouyon F and Madison G (2014) Syncopation (2013). Syncopation and the score. PLoS ONE 8:e74692. doi: 10.1371/jour- creates the sensation of groove in synthesized music examples. Front. Psychol. 5:1036. nal.pone.0074692 doi: 10.3389/fpsyg.2014.01036 Temperley, D. (1999). Syncopation in rock: a perceptual perspective. Pop. Music 18, This article was submitted to Auditory Cognitive Neuroscience, a section of the journal 19–40. doi: 10.1017/S0261143000008710 Frontiers in Psychology. Witek, M. A. G., Clarke, E. F., Wallentin, M., Kringelbach, M. L., and Vuust, P. Copyright © 2014 Sioros, Miron, Davies, Gouyon and Madison. This is an open- (2014). Syncopation, body-movement and pleasure in groove music. PLoS ONE access article distributed under the terms of the Creative Commons Attribution 9:e94446. doi: 10.1371/journal.pone.0094446 License (CC BY). The use, distribution or reproduction in other forums is permit- ted, provided the original author(s) or licensor are credited and that the original Conflict of Interest Statement: The authors declare that the research was con- publication in this journal is cited, in accordance with accepted academic practice. ducted in the absence of any commercial or financial relationships that could be No use, distribution or reproduction is permitted which does not comply with these construed as a potential conflict of interest. terms.

Frontiers in Psychology | Auditory Cognitive Neuroscience September 2014 | Volume 5 | Article 1036 | 10